(GURU 3D Rumor) Would you rather have two Nvidia X80Ti's? Or one X80 Titan?

[Dr. Zaus]

Real question is, who here gonna sell a kidney to afford x2 titanX?

I drive a car valued at 2% of my income... I don't see why I shouldn't have a gaming rig at the same value.

Words with Friends is going to be SO GOOD!

 

[Grubbernaught]

Two Titan-X's? There's a few around here who will snatch some up. I will wait for the 980ti replacement myself and see how things end up there. I don't think I'm capable of buying a single GPU, even though I tell myself I should. If I can have one badass GPU, having two is always far too tempting to pass up. Two pieces of awesomeness installed in my case is so much better than having only one.

Most definitely this. I've SLI'd every family since Fermi and I'm not sure I could stop if I wanted to.

This is possibly the gen I go triple cause there is no kill better than overkill.

 

[Compddd]

Fake leaked specs are fake.

 

[moonbogg]

Yeah I have been doing SLI since Nvidia started doing it with the Geforce 6 series. Won't be stopping now. If SLI supports drops off and it becomes a crappy option, the I'll stop. It would take a few generations though.

 

[RadiclDreamer]

I will always choose 1 card over 2

there are still far too many issues with SLI and crossfire to bother dealing with it

I am in the same boat man, nothing but tinkering and tweaking and still having a subpar experience with both SLI and Xfire

 

[ShintaiDK]

Always 1 card. SLI/CF is going to die out with DX12 as well.

And as said, its more trouble than worth already.

 

[moonbogg]

Well even if MGPU totally dies out, it had a good run. It was a much better run than in the 3DFX days. I think it will remain in some form or another. People want the performance. A single card is just lame. I'm used to the power of two now, but I admit that power is short lived. As soon as a new GPU is released it makes my two GPU's look like a stupid idea, but if you want more performance than is possible with a single card, there is no other choice.

 

[RadiclDreamer]

Well even if MGPU totally dies out, it had a good run. It was a much better run than in the 3DFX days. I think it will remain in some form or another. People want the performance. A single card is just lame. I'm used to the power of two now, but I admit that power is short lived. As soon as a new GPU is released it makes my two GPU's look like a stupid idea, but if you want more performance than is possible with a single card, there is no other choice.

I had far fewer issues with my voodoo2 SLI setup than with current gen, maybe im the oddball.

 

[exar333]

I had far fewer issues with my voodoo2 SLI setup than with current gen, maybe im the oddball.

SLI didn't require game profiles for 3DFX, if I remember right. It just worked...assuming you had identical cards.

 

[alcoholbob]

The real question is do you want to pay $1000+ for a 400-450mm² chip when a 550-600mm² Titan 3 will likely be out in end of early 2018 when yields have improved at TSMC? I mean that's a pretty poor value proposition.

With the GTX 1080 likely being a 250-300mm² chip, and the Titan card typically 50% bigger than the midrange, we are looking at a "small big die" of ~400mm² first generation most likely will be at best as fast as 980 Ti SLI, at worst it'll be 20-30% slower. Why even bother with the "upgrade" if you can wait another generation for the "Keplar of 16mm" (aka Volta) which will likely be a monster big die that'll make the first generation 16nm Titan look like a mid range chip in comparison?

Also since both have a 225W power envelope (if the specs are accurate) I suspect that there won't be that much a perf difference, especially since both will be thermally limited and TDP limited on a reference board. Titan X had 10% more cores than 980Ti and ended up 2% faster in practice with the stock BIOS, so if that same logic carries on, the Titan X-2 should be no more than 5% faster than the 1080 Ti under a stock BIOS if both are limited to a max of ~250W. It doesn't matter if you have 20% more cores if you have to run them at a lower frequency before hitting 80C/250W during sustained gaming.

 

[moonbogg]

The real question is do you want to pay $1000+ for a 400-450mm² chip when a 550-600mm² Titan 3 will likely be out in end of early 2018 when yields have improved at TSMC? I mean that's a pretty poor value proposition.

With the GTX 1080 likely being a 250-300mm² chip, and the Titan card typically 50% bigger than the midrange, we are looking at a "small big die" of ~400mm² first generation most likely will be at best as fast as 980 Ti SLI, at worst it'll be 20-30% slower. Why even bother with the "upgrade" if you can wait another generation for the "Keplar of 16mm" (aka Volta) which will likely be a monster big die that'll make the first generation 16nm Titan look like a mid range chip in comparison?

Also since both have a 225W power envelope (if the specs are accurate) I suspect that there won't be that much a perf difference, especially since both will be thermally limited and TDP limited on a reference board. Titan X had 10% more cores than 980Ti and ended up 2% faster in practice with the stock BIOS, so if that same logic carries on, the Titan X-2 should be no more than 5% faster than the 1080 Ti under a stock BIOS if both are limited to a max of ~250W. It doesn't matter if you have 20% more cores if you have to run them at a lower frequency before hitting 80C/250W during sustained gaming.

I see what you are saying. If we wait for Volta's big chip, that's one hell of a wait though. It could be like a half decade wait.

 

[Innokentij]

Never wait for hardware is my motto, just buy whatever is king for u, haters gonna hate anyway. If it makes u happy and u want it now u better get it now.

 

[Dr. Zaus]

Never wait for hardware is my motto, just buy whatever is king for u, haters gonna hate anyway. If it makes u happy and u want it now u better get it now.

Using the letter u in place of you really sells your sarcasm well.

 

[moonbogg]

Never wait for hardware is my motto, just buy whatever is king for u, haters gonna hate anyway. If it makes u happy and u want it now u better get it now.

Uh Uh Uh...Uh...BUY IT NOW BECHTAS!

 

[dark zero]

Guru3D forums birthed the following rumor. Lets assume its true. Based on the specs, what would you rather have? Two of the Ti's or a single Titan and OC it as far as possible? I'd still take my usual SLI route of the 2nd best card. I've been doing that forever except for like 1 round back with the 7800GTX's.

I hope that the cost doen't end being increasing like that.... is a nice boost from nVIDIA. Wondering if the 16nm process will be for all their cards, even their X30 and X10 cards

 

[ShintaiDK]

I had far fewer issues with my voodoo2 SLI setup than with current gen, maybe im the oddball.

The method is why. The old day SLI would have awful performance today because too much require the entire frame to be rendered. And not just half of it. On the other hand it worked flawless.

It was both the strength and weakness of the good old SLI.

 

[Innokentij]

Using the letter u in place of you really sells your sarcasm well.

No sarcasm, i meant what i wrote. :thumbsup:

 

[myocardia]

If Directx12 doesn't change that VR will. A GPU for each eye is probably the first natural use for SLI/CF.

SLI & CF are for sending two card's worth of data to one display. VR is about sending two signals that are different from each other to two different screens. While a high-end VR setup will have two GPUs, they won't/can't be SLI'd or CF'd, because the two GPUs will be producing different signals, the same way that your eyes do every day in real life.

 

[Grooveriding]

I was surprised this time how close the 980ti was to the Titan X. I'd think they'd want to distance them further to make the more expensive card appealing. I swapped my TX SLI for 980ti SLI because they were so close and I could get a little more performance out of aftermarket 980tis. I really like aftermarket cards over reference designs, but if the Titan card had enough extra that you couldn't get there with the ti, then I would go for Titans. I'd always prefer one card to two if I could get away with a single card. I use one card as often as possible unless I need both cards to get enough frames, there are only about 4 or 5 games where I've found the need for both cards.

I don't think we're going to get the Pascal Titan goods for awhile anyways. The rumours look like we're going to see a somewhat disappointing initial Pascal card. I think it may be as bad a proposition to TX/980ti users as the 980 was to 780ti owners when it launched. Maybe 10% better with games on the market when it launches, then they'll nerf Maxwell down in future games through driver neglect to make the Pascal card look better :sneaky:

 

[moonbogg]

I was surprised this time how close the 980ti was to the Titan X. I'd think they'd want to distance them further to make the more expensive card appealing. I swapped my TX SLI for 980ti SLI because they were so close and I could get a little more performance out of aftermarket 980tis. I really like aftermarket cards over reference designs, but if the Titan card had enough extra that you couldn't get there with the ti, then I would go for Titans. I'd always prefer one card to two if I could get away with a single card. I use one card as often as possible unless I need both cards to get enough frames, there are only about 4 or 5 games where I've found the need for both cards.

I don't think we're going to get the Pascal Titan goods for awhile anyways. The rumours look like we're going to see a somewhat disappointing initial Pascal card. I think it may be as bad a proposition to TX/980ti users as the 980 was to 780ti owners when it launched. Maybe 10% better with games on the market when it launches, then they'll nerf Maxwell down in future games through driver neglect to make the Pascal card look better :sneaky:

Correct. I'm gathering pitchforks and torches. When that day comes, we march.

 

[Qwertilot]

I'd lay off those pitch forks - if they can produce a 1080 card which is enough of a increase over the 980ti to tempt upgrades then they surely will. Its in their interest too

Might well not be possible of course. The ~2 times card definitely won't be until we get proper volume production of HBM2.

 

[Mahigan]

These specs aren't true at all. Take the GeForce X80 mentioned here. It has 128ROps and a 384-bit memory controller. Now how is that possible?

Take Kepler, it had a 384-bit memory controller and 48ROps. Each segment of 8ROps were tied to a single 64-bit memory controller segment. So the ROp:MC ratio was 8:1.

NVIDIA increased this to 16:1 with Maxwell. So each 16ROps were tied to a single 64-bit memory controller segment.

Now look at the specs claimed here for Pascal X80. 128ROps and 384-bit memory controller. Divide 384 by 64 and you get 6. Divide 128ROps by 6 and you get 21.3ROps per 64-bit memory controller segment. That's impossible thus invalidating these specs.

Now since Maxwell's ROps were already memory bandwidth starved at a 16:1 ratio, increasing this ratio would make no sense. So if we take the claimed X80 Ti specs they mention a 512-bit memory controller and 192ROps. If we take 512 and divide it by 64 we get 8. If we take 192ROps and divide by 8 we get 24. So what these specs claim is a 24:1 ROp to 64-bit memory controller ratio. That's impossible. If anything Pascal, on a 512-bit bus, would move to 128ROps if it retained its 16:1 ratio, not 192.

Bogus specs.

 

[3DVagabond]

What Titan has ever been worth it compared to a single *80 ti?

 

[Aristotelian]

About 'worth it' - I don't know. If they released a true Titan X successor this summer, I'd buy two and put them in SLI - reasonably flexible on price tbh. I'd do this understanding that the true Ti successor may take 1 year or more to emerge after that. If both were available at the same time, I'd go for two Tis probably because they are more 'gaming geared' I guess, and cheaper. But for me the Titan would be more 'worth it' because it comes out earlier - I think the Titan X was revealed at GDC 2015. They didn't reveal a Pascal Titan at GDC 2016, but where and when can we expect a serious pascal reveal in the near future? EDIT: it seems that the GTC on 4-7 April 2016 may include gpu unveiling.

Alcoholbob put forward a great question - "The real question is do you want to pay $1000+ for a 400-450mm² chip when a 550-600mm² Titan 3 will likely be out in end of early 2018 when yields have improved at TSMC? I mean that's a pretty poor value proposition" - and he's right. I don't think I understand the complexity of 'yield issues', but is their strategy really to focus first on small chips, make money off those, then move to the higher end larger chips because of 'yield' issues?

I thought Nvidia would do well by putting out a 500mm2 flagship titan and milking the gaming market with a 999 USD or more card for a while, then cutting that down for the cheaper versions. Anyone with upgrade-itis would do well to go for the flagship earlier. Value seeking customers are normally willing to wait longer. Perhaps I have it all wrong.

 

[xthetenth]

There's enough people who need the fastest and continually cycle out their cards to sell some of the biggest medium chip, but it's mostly there to serve as a price decoy so the medium crop sells like the 970.

I think their 500mm^2 chip will be available but given demand you'd better be prepared to fork out a few thousand for a version with no fan or video outputs. Those will sell enough and for a high enough price to cover yield issues. Basically think GK 100. Compute first, then finally Titan then Ti a while after that.